Three questions that keep coming up in our AI training

In AI training sessions with CERAGEM, Kurly, Dong-A ST, and the National Health Insurance Service, we heard the same three questions from teams in very different industries.

The first is, "Should we just build a chatbot?" There's a trap in that question. Building HeySeorak taught us that what a foreign traveler needs isn't one more chatbot repeating information, it's a conversion design that keeps them from dropping off at the door and gets them to an actual order. Before adding a tool, you have to see where the drop-off happens.

The second is, "We don't have data, how do we even start with AI?" mindtouch didn't start with data either. Both the non-face-to-face mental-health app pilot we ran with the city of Sokcho in 2021 and the HeySeorak beta running now started small and built the data ourselves as we went. Data isn't something you wait to have ready; it accumulates once you run a small pilot.

The third is, "We don't have a dedicated team for this." I've run three ventures by connecting planning, building, and operating myself, end to end. Not having a dedicated team isn't a reason you can't start. If anything, it's a reason to start smaller.

The most useful thing we can share in an AI session is a way to approach these questions. It comes from building and running services ourselves.